net/netfilter/nf_nat_core.c

   1 /*
   2  * (C) 1999-2001 Paul `Rusty' Russell
   3  * (C) 2002-2006 Netfilter Core Team <coreteam@netfilter.org>
   4  * (C) 2011 Patrick McHardy <kaber@trash.net>
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 as
   8  * published by the Free Software Foundation.
   9  */
  10
  11 #include <linux/module.h>
  12 #include <linux/types.h>
  13 #include <linux/timer.h>
  14 #include <linux/skbuff.h>
  15 #include <linux/gfp.h>
  16 #include <net/xfrm.h>
  17 #include <linux/jhash.h>
  18 #include <linux/rtnetlink.h>
  19
  20 #include <net/netfilter/nf_conntrack.h>
  21 #include <net/netfilter/nf_conntrack_core.h>
  22 #include <net/netfilter/nf_nat.h>
  23 #include <net/netfilter/nf_nat_l3proto.h>
  24 #include <net/netfilter/nf_nat_l4proto.h>
  25 #include <net/netfilter/nf_nat_core.h>
  26 #include <net/netfilter/nf_nat_helper.h>
  27 #include <net/netfilter/nf_conntrack_helper.h>
  28 #include <net/netfilter/nf_conntrack_seqadj.h>
  29 #include <net/netfilter/nf_conntrack_l3proto.h>
  30 #include <net/netfilter/nf_conntrack_zones.h>
  31 #include <linux/netfilter/nf_nat.h>
  32
  33 static DEFINE_SPINLOCK(nf_nat_lock);
  34
  35 static DEFINE_MUTEX(nf_nat_proto_mutex);
  36 static const struct nf_nat_l3proto __rcu *nf_nat_l3protos[NFPROTO_NUMPROTO]
  37                                                 __read_mostly;
  38 static const struct nf_nat_l4proto __rcu **nf_nat_l4protos[NFPROTO_NUMPROTO]
  39                                                 __read_mostly;
  40
  41
  42 inline const struct nf_nat_l3proto *
  43 __nf_nat_l3proto_find(u8 family)
  44 {
  45         return rcu_dereference(nf_nat_l3protos[family]);
  46 }
  47
  48 inline const struct nf_nat_l4proto *
  49 __nf_nat_l4proto_find(u8 family, u8 protonum)
  50 {
  51         return rcu_dereference(nf_nat_l4protos[family][protonum]);
  52 }
  53 EXPORT_SYMBOL_GPL(__nf_nat_l4proto_find);
  54
  55 #ifdef CONFIG_XFRM
  56 static void __nf_nat_decode_session(struct sk_buff *skb, struct flowi *fl)
  57 {
  58         const struct nf_nat_l3proto *l3proto;
  59         const struct nf_conn *ct;
  60         enum ip_conntrack_info ctinfo;
  61         enum ip_conntrack_dir dir;
  62         unsigned  long statusbit;
  63         u8 family;
  64
  65         ct = nf_ct_get(skb, &ctinfo);
  66         if (ct == NULL)
  67                 return;
  68
  69         family = ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple.src.l3num;
  70         rcu_read_lock();
  71         l3proto = __nf_nat_l3proto_find(family);
  72         if (l3proto == NULL)
  73                 goto out;
  74
  75         dir = CTINFO2DIR(ctinfo);
  76         if (dir == IP_CT_DIR_ORIGINAL)
  77                 statusbit = IPS_DST_NAT;
  78         else
  79                 statusbit = IPS_SRC_NAT;
  80
  81         l3proto->decode_session(skb, ct, dir, statusbit, fl);
  82 out:
  83         rcu_read_unlock();
  84 }
  85
  86 int nf_xfrm_me_harder(struct sk_buff *skb, unsigned int family)
  87 {
  88         struct flowi fl;
  89         unsigned int hh_len;
  90         struct dst_entry *dst;
  91         int err;
  92
  93         err = xfrm_decode_session(skb, &fl, family);
  94         if (err < 0)
  95                 return err;
  96
  97         dst = skb_dst(skb);
  98         if (dst->xfrm)
  99                 dst = ((struct xfrm_dst *)dst)->route;
 100         dst_hold(dst);
 101
 102         dst = xfrm_lookup(dev_net(dst->dev), dst, &fl, skb->sk, 0);
 103         if (IS_ERR(dst))
 104                 return PTR_ERR(dst);
 105
 106         skb_dst_drop(skb);
 107         skb_dst_set(skb, dst);
 108
 109         /* Change in oif may mean change in hh_len. */
 110         hh_len = skb_dst(skb)->dev->hard_header_len;
 111         if (skb_headroom(skb) < hh_len &&
 112             pskb_expand_head(skb, hh_len - skb_headroom(skb), 0, GFP_ATOMIC))
 113                 return -ENOMEM;
 114         return 0;
 115 }
 116 EXPORT_SYMBOL(nf_xfrm_me_harder);
 117 #endif /* CONFIG_XFRM */
 118
 119 /* We keep an extra hash for each conntrack, for fast searching. */
 120 static inline unsigned int
 121 hash_by_src(const struct net *net, u16 zone,
 122             const struct nf_conntrack_tuple *tuple)
 123 {
 124         unsigned int hash;
 125
 126         /* Original src, to ensure we map it consistently if poss. */
 127         hash = jhash2((u32 *)&tuple->src, sizeof(tuple->src) / sizeof(u32),
 128                       tuple->dst.protonum ^ zone ^ nf_conntrack_hash_rnd);
 129         return ((u64)hash * net->ct.nat_htable_size) >> 32;
 130 }
 131
 132 /* Is this tuple already taken? (not by us) */
 133 int
 134 nf_nat_used_tuple(const struct nf_conntrack_tuple *tuple,
 135                   const struct nf_conn *ignored_conntrack)
 136 {
 137         /* Conntrack tracking doesn't keep track of outgoing tuples; only
 138          * incoming ones.  NAT means they don't have a fixed mapping,
 139          * so we invert the tuple and look for the incoming reply.
 140          *
 141          * We could keep a separate hash if this proves too slow.
 142          */
 143         struct nf_conntrack_tuple reply;
 144
 145         nf_ct_invert_tuplepr(&reply, tuple);
 146         return nf_conntrack_tuple_taken(&reply, ignored_conntrack);
 147 }
 148 EXPORT_SYMBOL(nf_nat_used_tuple);
 149
 150 /* If we source map this tuple so reply looks like reply_tuple, will
 151  * that meet the constraints of range.
 152  */
 153 static int in_range(const struct nf_nat_l3proto *l3proto,
 154                     const struct nf_nat_l4proto *l4proto,
 155                     const struct nf_conntrack_tuple *tuple,
 156                     const struct nf_nat_range *range)
 157 {
 158         /* If we are supposed to map IPs, then we must be in the
 159          * range specified, otherwise let this drag us onto a new src IP.
 160          */
 161         if (range->flags & NF_NAT_RANGE_MAP_IPS &&
 162             !l3proto->in_range(tuple, range))
 163                 return 0;
 164
 165         if (!(range->flags & NF_NAT_RANGE_PROTO_SPECIFIED) ||
 166             l4proto->in_range(tuple, NF_NAT_MANIP_SRC,
 167                               &range->min_proto, &range->max_proto))
 168                 return 1;
 169
 170         return 0;
 171 }
 172
 173 static inline int
 174 same_src(const struct nf_conn *ct,
 175          const struct nf_conntrack_tuple *tuple)
 176 {
 177         const struct nf_conntrack_tuple *t;
 178
 179         t = &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple;
 180         return (t->dst.protonum == tuple->dst.protonum &&
 181                 nf_inet_addr_cmp(&t->src.u3, &tuple->src.u3) &&
 182                 t->src.u.all == tuple->src.u.all);
 183 }
 184
 185 /* Only called for SRC manip */
 186 static int
 187 find_appropriate_src(struct net *net, u16 zone,
 188                      const struct nf_nat_l3proto *l3proto,
 189                      const struct nf_nat_l4proto *l4proto,
 190                      const struct nf_conntrack_tuple *tuple,
 191                      struct nf_conntrack_tuple *result,
 192                      const struct nf_nat_range *range)
 193 {
 194         unsigned int h = hash_by_src(net, zone, tuple);
 195         const struct nf_conn_nat *nat;
 196         const struct nf_conn *ct;
 197
 198         hlist_for_each_entry_rcu(nat, &net->ct.nat_bysource[h], bysource) {
 199                 ct = nat->ct;
 200                 if (same_src(ct, tuple) && nf_ct_zone(ct) == zone) {
 201                         /* Copy source part from reply tuple. */
 202                         nf_ct_invert_tuplepr(result,
 203                                        &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 204                         result->dst = tuple->dst;
 205
 206                         if (in_range(l3proto, l4proto, result, range))
 207                                 return 1;
 208                 }
 209         }
 210         return 0;
 211 }
 212
 213 /* For [FUTURE] fragmentation handling, we want the least-used
 214  * src-ip/dst-ip/proto triple.  Fairness doesn't come into it.  Thus
 215  * if the range specifies 1.2.3.4 ports 10000-10005 and 1.2.3.5 ports
 216  * 1-65535, we don't do pro-rata allocation based on ports; we choose
 217  * the ip with the lowest src-ip/dst-ip/proto usage.
 218  */
 219 static void
 220 find_best_ips_proto(u16 zone, struct nf_conntrack_tuple *tuple,
 221                     const struct nf_nat_range *range,
 222                     const struct nf_conn *ct,
 223                     enum nf_nat_manip_type maniptype)
 224 {
 225         union nf_inet_addr *var_ipp;
 226         unsigned int i, max;
 227         /* Host order */
 228         u32 minip, maxip, j, dist;
 229         bool full_range;
 230
 231         /* No IP mapping?  Do nothing. */
 232         if (!(range->flags & NF_NAT_RANGE_MAP_IPS))
 233                 return;
 234
 235         if (maniptype == NF_NAT_MANIP_SRC)
 236                 var_ipp = &tuple->src.u3;
 237         else
 238                 var_ipp = &tuple->dst.u3;
 239
 240         /* Fast path: only one choice. */
 241         if (nf_inet_addr_cmp(&range->min_addr, &range->max_addr)) {
 242                 *var_ipp = range->min_addr;
 243                 return;
 244         }
 245
 246         if (nf_ct_l3num(ct) == NFPROTO_IPV4)
 247                 max = sizeof(var_ipp->ip) / sizeof(u32) - 1;
 248         else
 249                 max = sizeof(var_ipp->ip6) / sizeof(u32) - 1;
 250
 251         /* Hashing source and destination IPs gives a fairly even
 252          * spread in practice (if there are a small number of IPs
 253          * involved, there usually aren't that many connections
 254          * anyway).  The consistency means that servers see the same
 255          * client coming from the same IP (some Internet Banking sites
 256          * like this), even across reboots.
 257          */
 258         j = jhash2((u32 *)&tuple->src.u3, sizeof(tuple->src.u3) / sizeof(u32),
 259                    range->flags & NF_NAT_RANGE_PERSISTENT ?
 260                         0 : (__force u32)tuple->dst.u3.all[max] ^ zone);
 261
 262         full_range = false;
 263         for (i = 0; i <= max; i++) {
 264                 /* If first bytes of the address are at the maximum, use the
 265                  * distance. Otherwise use the full range.
 266                  */
 267                 if (!full_range) {
 268                         minip = ntohl((__force __be32)range->min_addr.all[i]);
 269                         maxip = ntohl((__force __be32)range->max_addr.all[i]);
 270                         dist  = maxip - minip + 1;
 271                 } else {
 272                         minip = 0;
 273                         dist  = ~0;
 274                 }
 275
 276                 var_ipp->all[i] = (__force __u32)
 277                         htonl(minip + (((u64)j * dist) >> 32));
 278                 if (var_ipp->all[i] != range->max_addr.all[i])
 279                         full_range = true;
 280
 281                 if (!(range->flags & NF_NAT_RANGE_PERSISTENT))
 282                         j ^= (__force u32)tuple->dst.u3.all[i];
 283         }
 284 }
 285
 286 /* Manipulate the tuple into the range given. For NF_INET_POST_ROUTING,
 287  * we change the source to map into the range. For NF_INET_PRE_ROUTING
 288  * and NF_INET_LOCAL_OUT, we change the destination to map into the
 289  * range. It might not be possible to get a unique tuple, but we try.
 290  * At worst (or if we race), we will end up with a final duplicate in
 291  * __ip_conntrack_confirm and drop the packet. */
 292 static void
 293 get_unique_tuple(struct nf_conntrack_tuple *tuple,
 294                  const struct nf_conntrack_tuple *orig_tuple,
 295                  const struct nf_nat_range *range,
 296                  struct nf_conn *ct,
 297                  enum nf_nat_manip_type maniptype)
 298 {
 299         const struct nf_nat_l3proto *l3proto;
 300         const struct nf_nat_l4proto *l4proto;
 301         struct net *net = nf_ct_net(ct);
 302         u16 zone = nf_ct_zone(ct);
 303
 304         rcu_read_lock();
 305         l3proto = __nf_nat_l3proto_find(orig_tuple->src.l3num);
 306         l4proto = __nf_nat_l4proto_find(orig_tuple->src.l3num,
 307                                         orig_tuple->dst.protonum);
 308
 309         /* 1) If this srcip/proto/src-proto-part is currently mapped,
 310          * and that same mapping gives a unique tuple within the given
 311          * range, use that.
 312          *
 313          * This is only required for source (ie. NAT/masq) mappings.
 314          * So far, we don't do local source mappings, so multiple
 315          * manips not an issue.
 316          */
 317         if (maniptype == NF_NAT_MANIP_SRC &&
 318             !(range->flags & NF_NAT_RANGE_PROTO_RANDOM)) {
 319                 /* try the original tuple first */
 320                 if (in_range(l3proto, l4proto, orig_tuple, range)) {
 321                         if (!nf_nat_used_tuple(orig_tuple, ct)) {
 322                                 *tuple = *orig_tuple;
 323                                 goto out;
 324                         }
 325                 } else if (find_appropriate_src(net, zone, l3proto, l4proto,
 326                                                 orig_tuple, tuple, range)) {
 327                         pr_debug("get_unique_tuple: Found current src map\n");
 328                         if (!nf_nat_used_tuple(tuple, ct))
 329                                 goto out;
 330                 }
 331         }
 332
 333         /* 2) Select the least-used IP/proto combination in the given range */
 334         *tuple = *orig_tuple;
 335         find_best_ips_proto(zone, tuple, range, ct, maniptype);
 336
 337         /* 3) The per-protocol part of the manip is made to map into
 338          * the range to make a unique tuple.
 339          */
 340
 341         /* Only bother mapping if it's not already in range and unique */
 342         if (!(range->flags & NF_NAT_RANGE_PROTO_RANDOM)) {
 343                 if (range->flags & NF_NAT_RANGE_PROTO_SPECIFIED) {
 344                         if (l4proto->in_range(tuple, maniptype,
 345                                               &range->min_proto,
 346                                               &range->max_proto) &&
 347                             (range->min_proto.all == range->max_proto.all ||
 348                              !nf_nat_used_tuple(tuple, ct)))
 349                                 goto out;
 350                 } else if (!nf_nat_used_tuple(tuple, ct)) {
 351                         goto out;
 352                 }
 353         }
 354
 355         /* Last change: get protocol to try to obtain unique tuple. */
 356         l4proto->unique_tuple(l3proto, tuple, range, maniptype, ct);
 357 out:
 358         rcu_read_unlock();
 359 }
 360
 361 unsigned int
 362 nf_nat_setup_info(struct nf_conn *ct,
 363                   const struct nf_nat_range *range,
 364                   enum nf_nat_manip_type maniptype)
 365 {
 366         struct net *net = nf_ct_net(ct);
 367         struct nf_conntrack_tuple curr_tuple, new_tuple;
 368         struct nf_conn_nat *nat;
 369
 370         /* nat helper or nfctnetlink also setup binding */
 371         nat = nfct_nat(ct);
 372         if (!nat) {
 373                 nat = nf_ct_ext_add(ct, NF_CT_EXT_NAT, GFP_ATOMIC);
 374                 if (nat == NULL) {
 375                         pr_debug("failed to add NAT extension\n");
 376                         return NF_ACCEPT;
 377                 }
 378         }
 379
 380         NF_CT_ASSERT(maniptype == NF_NAT_MANIP_SRC ||
 381                      maniptype == NF_NAT_MANIP_DST);
 382         BUG_ON(nf_nat_initialized(ct, maniptype));
 383
 384         /* What we've got will look like inverse of reply. Normally
 385          * this is what is in the conntrack, except for prior
 386          * manipulations (future optimization: if num_manips == 0,
 387          * orig_tp = ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple)
 388          */
 389         nf_ct_invert_tuplepr(&curr_tuple,
 390                              &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 391
 392         get_unique_tuple(&new_tuple, &curr_tuple, range, ct, maniptype);
 393
 394         if (!nf_ct_tuple_equal(&new_tuple, &curr_tuple)) {
 395                 struct nf_conntrack_tuple reply;
 396
 397                 /* Alter conntrack table so will recognize replies. */
 398                 nf_ct_invert_tuplepr(&reply, &new_tuple);
 399                 nf_conntrack_alter_reply(ct, &reply);
 400
 401                 /* Non-atomic: we own this at the moment. */
 402                 if (maniptype == NF_NAT_MANIP_SRC)
 403                         ct->status |= IPS_SRC_NAT;
 404                 else
 405                         ct->status |= IPS_DST_NAT;
 406
 407                 if (nfct_help(ct))
 408                         nfct_seqadj_ext_add(ct);
 409         }
 410
 411         if (maniptype == NF_NAT_MANIP_SRC) {
 412                 unsigned int srchash;
 413
 414                 srchash = hash_by_src(net, nf_ct_zone(ct),
 415                                       &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
 416                 spin_lock_bh(&nf_nat_lock);
 417                 /* nf_conntrack_alter_reply might re-allocate extension aera */
 418                 nat = nfct_nat(ct);
 419                 nat->ct = ct;
 420                 hlist_add_head_rcu(&nat->bysource,
 421                                    &net->ct.nat_bysource[srchash]);
 422                 spin_unlock_bh(&nf_nat_lock);
 423         }
 424
 425         /* It's done. */
 426         if (maniptype == NF_NAT_MANIP_DST)
 427                 ct->status |= IPS_DST_NAT_DONE;
 428         else
 429                 ct->status |= IPS_SRC_NAT_DONE;
 430
 431         return NF_ACCEPT;
 432 }
 433 EXPORT_SYMBOL(nf_nat_setup_info);
 434
 435 /* Do packet manipulations according to nf_nat_setup_info. */
 436 unsigned int nf_nat_packet(struct nf_conn *ct,
 437                            enum ip_conntrack_info ctinfo,
 438                            unsigned int hooknum,
 439                            struct sk_buff *skb)
 440 {
 441         const struct nf_nat_l3proto *l3proto;
 442         const struct nf_nat_l4proto *l4proto;
 443         enum ip_conntrack_dir dir = CTINFO2DIR(ctinfo);
 444         unsigned long statusbit;
 445         enum nf_nat_manip_type mtype = HOOK2MANIP(hooknum);
 446
 447         if (mtype == NF_NAT_MANIP_SRC)
 448                 statusbit = IPS_SRC_NAT;
 449         else
 450                 statusbit = IPS_DST_NAT;
 451
 452         /* Invert if this is reply dir. */
 453         if (dir == IP_CT_DIR_REPLY)
 454                 statusbit ^= IPS_NAT_MASK;
 455
 456         /* Non-atomic: these bits don't change. */
 457         if (ct->status & statusbit) {
 458                 struct nf_conntrack_tuple target;
 459
 460                 /* We are aiming to look like inverse of other direction. */
 461                 nf_ct_invert_tuplepr(&target, &ct->tuplehash[!dir].tuple);
 462
 463                 l3proto = __nf_nat_l3proto_find(target.src.l3num);
 464                 l4proto = __nf_nat_l4proto_find(target.src.l3num,
 465                                                 target.dst.protonum);
 466                 if (!l3proto->manip_pkt(skb, 0, l4proto, &target, mtype))
 467                         return NF_DROP;
 468         }
 469         return NF_ACCEPT;
 470 }
 471 EXPORT_SYMBOL_GPL(nf_nat_packet);
 472
 473 struct nf_nat_proto_clean {
 474         u8      l3proto;
 475         u8      l4proto;
 476 };
 477
 478 /* kill conntracks with affected NAT section */
 479 static int nf_nat_proto_remove(struct nf_conn *i, void *data)
 480 {
 481         const struct nf_nat_proto_clean *clean = data;
 482         struct nf_conn_nat *nat = nfct_nat(i);
 483
 484         if (!nat)
 485                 return 0;
 486
 487         if ((clean->l3proto && nf_ct_l3num(i) != clean->l3proto) ||
 488             (clean->l4proto && nf_ct_protonum(i) != clean->l4proto))
 489                 return 0;
 490
 491         return i->status & IPS_NAT_MASK ? 1 : 0;
 492 }
 493
 494 static void nf_nat_l4proto_clean(u8 l3proto, u8 l4proto)
 495 {
 496         struct nf_nat_proto_clean clean = {
 497                 .l3proto = l3proto,
 498                 .l4proto = l4proto,
 499         };
 500         struct net *net;
 501
 502         rtnl_lock();
 503         for_each_net(net)
 504                 nf_ct_iterate_cleanup(net, nf_nat_proto_remove, &clean, 0, 0);
 505         rtnl_unlock();
 506 }
 507
 508 static void nf_nat_l3proto_clean(u8 l3proto)
 509 {
 510         struct nf_nat_proto_clean clean = {
 511                 .l3proto = l3proto,
 512         };
 513         struct net *net;
 514
 515         rtnl_lock();
 516
 517         for_each_net(net)
 518                 nf_ct_iterate_cleanup(net, nf_nat_proto_remove, &clean, 0, 0);
 519         rtnl_unlock();
 520 }
 521
 522 /* Protocol registration. */
 523 int nf_nat_l4proto_register(u8 l3proto, const struct nf_nat_l4proto *l4proto)
 524 {
 525         const struct nf_nat_l4proto **l4protos;
 526         unsigned int i;
 527         int ret = 0;
 528
 529         mutex_lock(&nf_nat_proto_mutex);
 530         if (nf_nat_l4protos[l3proto] == NULL) {
 531                 l4protos = kmalloc(IPPROTO_MAX * sizeof(struct nf_nat_l4proto *),
 532                                    GFP_KERNEL);
 533                 if (l4protos == NULL) {
 534                         ret = -ENOMEM;
 535                         goto out;
 536                 }
 537
 538                 for (i = 0; i < IPPROTO_MAX; i++)
 539                         RCU_INIT_POINTER(l4protos[i], &nf_nat_l4proto_unknown);
 540
 541                 /* Before making proto_array visible to lockless readers,
 542                  * we must make sure its content is committed to memory.
 543                  */
 544                 smp_wmb();
 545
 546                 nf_nat_l4protos[l3proto] = l4protos;
 547         }
 548
 549         if (rcu_dereference_protected(
 550                         nf_nat_l4protos[l3proto][l4proto->l4proto],
 551                         lockdep_is_held(&nf_nat_proto_mutex)
 552                         ) != &nf_nat_l4proto_unknown) {
 553                 ret = -EBUSY;
 554                 goto out;
 555         }
 556         RCU_INIT_POINTER(nf_nat_l4protos[l3proto][l4proto->l4proto], l4proto);
 557  out:
 558         mutex_unlock(&nf_nat_proto_mutex);
 559         return ret;
 560 }
 561 EXPORT_SYMBOL_GPL(nf_nat_l4proto_register);
 562
 563 /* No one stores the protocol anywhere; simply delete it. */
 564 void nf_nat_l4proto_unregister(u8 l3proto, const struct nf_nat_l4proto *l4proto)
 565 {
 566         mutex_lock(&nf_nat_proto_mutex);
 567         RCU_INIT_POINTER(nf_nat_l4protos[l3proto][l4proto->l4proto],
 568                          &nf_nat_l4proto_unknown);
 569         mutex_unlock(&nf_nat_proto_mutex);
 570         synchronize_rcu();
 571
 572         nf_nat_l4proto_clean(l3proto, l4proto->l4proto);
 573 }
 574 EXPORT_SYMBOL_GPL(nf_nat_l4proto_unregister);
 575
 576 int nf_nat_l3proto_register(const struct nf_nat_l3proto *l3proto)
 577 {
 578         int err;
 579
 580         err = nf_ct_l3proto_try_module_get(l3proto->l3proto);
 581         if (err < 0)
 582                 return err;
 583
 584         mutex_lock(&nf_nat_proto_mutex);
 585         RCU_INIT_POINTER(nf_nat_l4protos[l3proto->l3proto][IPPROTO_TCP],
 586                          &nf_nat_l4proto_tcp);
 587         RCU_INIT_POINTER(nf_nat_l4protos[l3proto->l3proto][IPPROTO_UDP],
 588                          &nf_nat_l4proto_udp);
 589         mutex_unlock(&nf_nat_proto_mutex);
 590
 591         RCU_INIT_POINTER(nf_nat_l3protos[l3proto->l3proto], l3proto);
 592         return 0;
 593 }
 594 EXPORT_SYMBOL_GPL(nf_nat_l3proto_register);
 595
 596 void nf_nat_l3proto_unregister(const struct nf_nat_l3proto *l3proto)
 597 {
 598         mutex_lock(&nf_nat_proto_mutex);
 599         RCU_INIT_POINTER(nf_nat_l3protos[l3proto->l3proto], NULL);
 600         mutex_unlock(&nf_nat_proto_mutex);
 601         synchronize_rcu();
 602
 603         nf_nat_l3proto_clean(l3proto->l3proto);
 604         nf_ct_l3proto_module_put(l3proto->l3proto);
 605 }
 606 EXPORT_SYMBOL_GPL(nf_nat_l3proto_unregister);
 607
 608 /* No one using conntrack by the time this called. */
 609 static void nf_nat_cleanup_conntrack(struct nf_conn *ct)
 610 {
 611         struct nf_conn_nat *nat = nf_ct_ext_find(ct, NF_CT_EXT_NAT);
 612
 613         if (nat == NULL || nat->ct == NULL)
 614                 return;
 615
 616         NF_CT_ASSERT(nat->ct->status & IPS_SRC_NAT_DONE);
 617
 618         spin_lock_bh(&nf_nat_lock);
 619         hlist_del_rcu(&nat->bysource);
 620         spin_unlock_bh(&nf_nat_lock);
 621 }
 622
 623 static void nf_nat_move_storage(void *new, void *old)
 624 {
 625         struct nf_conn_nat *new_nat = new;
 626         struct nf_conn_nat *old_nat = old;
 627         struct nf_conn *ct = old_nat->ct;
 628
 629         if (!ct || !(ct->status & IPS_SRC_NAT_DONE))
 630                 return;
 631
 632         spin_lock_bh(&nf_nat_lock);
 633         hlist_replace_rcu(&old_nat->bysource, &new_nat->bysource);
 634         spin_unlock_bh(&nf_nat_lock);
 635 }
 636
 637 static struct nf_ct_ext_type nat_extend __read_mostly = {
 638         .len            = sizeof(struct nf_conn_nat),
 639         .align          = __alignof__(struct nf_conn_nat),
 640         .destroy        = nf_nat_cleanup_conntrack,
 641         .move           = nf_nat_move_storage,
 642         .id             = NF_CT_EXT_NAT,
 643         .flags          = NF_CT_EXT_F_PREALLOC,
 644 };
 645
 646 #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE)
 647
 648 #include <linux/netfilter/nfnetlink.h>
 649 #include <linux/netfilter/nfnetlink_conntrack.h>
 650
 651 static const struct nla_policy protonat_nla_policy[CTA_PROTONAT_MAX+1] = {
 652         [CTA_PROTONAT_PORT_MIN] = { .type = NLA_U16 },
 653         [CTA_PROTONAT_PORT_MAX] = { .type = NLA_U16 },
 654 };
 655
 656 static int nfnetlink_parse_nat_proto(struct nlattr *attr,
 657                                      const struct nf_conn *ct,
 658                                      struct nf_nat_range *range)
 659 {
 660         struct nlattr *tb[CTA_PROTONAT_MAX+1];
 661         const struct nf_nat_l4proto *l4proto;
 662         int err;
 663
 664         err = nla_parse_nested(tb, CTA_PROTONAT_MAX, attr, protonat_nla_policy);
 665         if (err < 0)
 666                 return err;
 667
 668         l4proto = __nf_nat_l4proto_find(nf_ct_l3num(ct), nf_ct_protonum(ct));
 669         if (l4proto->nlattr_to_range)
 670                 err = l4proto->nlattr_to_range(tb, range);
 671
 672         return err;
 673 }
 674
 675 static const struct nla_policy nat_nla_policy[CTA_NAT_MAX+1] = {
 676         [CTA_NAT_V4_MINIP]      = { .type = NLA_U32 },
 677         [CTA_NAT_V4_MAXIP]      = { .type = NLA_U32 },
 678         [CTA_NAT_V6_MINIP]      = { .len = sizeof(struct in6_addr) },
 679         [CTA_NAT_V6_MAXIP]      = { .len = sizeof(struct in6_addr) },
 680         [CTA_NAT_PROTO]         = { .type = NLA_NESTED },
 681 };
 682
 683 static int
 684 nfnetlink_parse_nat(const struct nlattr *nat,
 685                     const struct nf_conn *ct, struct nf_nat_range *range)
 686 {
 687         const struct nf_nat_l3proto *l3proto;
 688         struct nlattr *tb[CTA_NAT_MAX+1];
 689         int err;
 690
 691         memset(range, 0, sizeof(*range));
 692
 693         err = nla_parse_nested(tb, CTA_NAT_MAX, nat, nat_nla_policy);
 694         if (err < 0)
 695                 return err;
 696
 697         rcu_read_lock();
 698         l3proto = __nf_nat_l3proto_find(nf_ct_l3num(ct));
 699         if (l3proto == NULL) {
 700                 err = -EAGAIN;
 701                 goto out;
 702         }
 703         err = l3proto->nlattr_to_range(tb, range);
 704         if (err < 0)
 705                 goto out;
 706
 707         if (!tb[CTA_NAT_PROTO])
 708                 goto out;
 709
 710         err = nfnetlink_parse_nat_proto(tb[CTA_NAT_PROTO], ct, range);
 711 out:
 712         rcu_read_unlock();
 713         return err;
 714 }
 715
 716 static int
 717 nfnetlink_parse_nat_setup(struct nf_conn *ct,
 718                           enum nf_nat_manip_type manip,
 719                           const struct nlattr *attr)
 720 {
 721         struct nf_nat_range range;
 722         int err;
 723
 724         err = nfnetlink_parse_nat(attr, ct, &range);
 725         if (err < 0)
 726                 return err;
 727         if (nf_nat_initialized(ct, manip))
 728                 return -EEXIST;
 729
 730         return nf_nat_setup_info(ct, &range, manip);
 731 }
 732 #else
 733 static int
 734 nfnetlink_parse_nat_setup(struct nf_conn *ct,
 735                           enum nf_nat_manip_type manip,
 736                           const struct nlattr *attr)
 737 {
 738         return -EOPNOTSUPP;
 739 }
 740 #endif
 741
 742 static int __net_init nf_nat_net_init(struct net *net)
 743 {
 744         /* Leave them the same for the moment. */
 745         net->ct.nat_htable_size = net->ct.htable_size;
 746         net->ct.nat_bysource = nf_ct_alloc_hashtable(&net->ct.nat_htable_size, 0);
 747         if (!net->ct.nat_bysource)
 748                 return -ENOMEM;
 749         return 0;
 750 }
 751
 752 static void __net_exit nf_nat_net_exit(struct net *net)
 753 {
 754         struct nf_nat_proto_clean clean = {};
 755
 756         nf_ct_iterate_cleanup(net, &nf_nat_proto_remove, &clean, 0, 0);
 757         synchronize_rcu();
 758         nf_ct_free_hashtable(net->ct.nat_bysource, net->ct.nat_htable_size);
 759 }
 760
 761 static struct pernet_operations nf_nat_net_ops = {
 762         .init = nf_nat_net_init,
 763         .exit = nf_nat_net_exit,
 764 };
 765
 766 static struct nf_ct_helper_expectfn follow_master_nat = {
 767         .name           = "nat-follow-master",
 768         .expectfn       = nf_nat_follow_master,
 769 };
 770
 771 static int __init nf_nat_init(void)
 772 {
 773         int ret;
 774
 775         ret = nf_ct_extend_register(&nat_extend);
 776         if (ret < 0) {
 777                 printk(KERN_ERR "nf_nat_core: Unable to register extension\n");
 778                 return ret;
 779         }
 780
 781         ret = register_pernet_subsys(&nf_nat_net_ops);
 782         if (ret < 0)
 783                 goto cleanup_extend;
 784
 785         nf_ct_helper_expectfn_register(&follow_master_nat);
 786
 787         /* Initialize fake conntrack so that NAT will skip it */
 788         nf_ct_untracked_status_or(IPS_NAT_DONE_MASK);
 789
 790         BUG_ON(nfnetlink_parse_nat_setup_hook != NULL);
 791         RCU_INIT_POINTER(nfnetlink_parse_nat_setup_hook,
 792                            nfnetlink_parse_nat_setup);
 793 #ifdef CONFIG_XFRM
 794         BUG_ON(nf_nat_decode_session_hook != NULL);
 795         RCU_INIT_POINTER(nf_nat_decode_session_hook, __nf_nat_decode_session);
 796 #endif
 797         return 0;
 798
 799  cleanup_extend:
 800         nf_ct_extend_unregister(&nat_extend);
 801         return ret;
 802 }
 803
 804 static void __exit nf_nat_cleanup(void)
 805 {
 806         unsigned int i;
 807
 808         unregister_pernet_subsys(&nf_nat_net_ops);
 809         nf_ct_extend_unregister(&nat_extend);
 810         nf_ct_helper_expectfn_unregister(&follow_master_nat);
 811         RCU_INIT_POINTER(nfnetlink_parse_nat_setup_hook, NULL);
 812 #ifdef CONFIG_XFRM
 813         RCU_INIT_POINTER(nf_nat_decode_session_hook, NULL);
 814 #endif
 815         for (i = 0; i < NFPROTO_NUMPROTO; i++)
 816                 kfree(nf_nat_l4protos[i]);
 817         synchronize_net();
 818 }
 819
 820 MODULE_LICENSE("GPL");
 821
 822 module_init(nf_nat_init);
 823 module_exit(nf_nat_cleanup);